How to assess AI skills in non-technical roles: 2 exercises

Researched 30 October 2026 · By Screened

Give the candidate a realistic work sample from the job. Score the output and the checks they make.

These two exercises show whether someone can create useful work with AI, catch unsupported claims and recognise when a human decision is still needed.

Use Exercise 1 on its own for roles where reviewing AI-assisted work is enough. Add Exercise 2 when the person will use an AI assistant directly.

What should you assess?

Ask the hiring manager:

Which task should this person be able to complete with AI, and what must they still check themselves?

For a recruiting coordinator, that might be preparing candidate updates. For an operations role, it might be drafting a handover. For a finance role, it might be checking a calculation or explaining a variance.

Turn the answer into an observable requirement:

Prepares candidate updates with AI and checks the final message against the actual hiring decision.

The BCS AI Skills and Adoption Report 2026 makes a similar distinction between technical AI development and everyday AI use and judgement. Its survey covered BCS members, rather than the wider UK workforce.

Does a specific AI product matter?

Product experience matters when the person must configure or maintain an existing system from day one.

For routine AI-assisted work, give everyone the same introduction to the approved tool. Then assess the quality of the work and the checks they make.

Before you run either exercise

• Try the task with colleagues before setting the time limit.
• Tell candidates the expected effort and permitted tools.
• Explain how to request an adjustment.
• Give everyone the same brief and the same introduction to the tool.
• Keep the exercise outside live hiring systems and do not require a paid personal subscription.
• Measure speed only when speed matters to the job.

Acas recommends preparation and appropriate adjustments for interviews. See its interview guidance.

With a live assistant, one candidate may receive a better first draft than another. Score the final work and the verification process, not how much editing the model happened to require.

How do you adapt these exercises to another role?

Keep three ingredients:

1. Source facts the candidate must preserve.
2. An unresolved decision they must recognise rather than invent.
3. A real output someone in the role would actually produce.

For an operations role, that output might be a handover. For finance, it might be a variance explanation. For customer support, it might be a reply to a complaint. Change the work, while keeping the assessment structure consistent.

Exercise 1: spot errors in an AI-written candidate email

Tell the candidate they are a recruiting coordinator reviewing an AI-generated email to an applicant. They should correct the draft and explain the important changes. They should not send anything.

Complete this exercise without generative AI, with any agreed accessibility support.

Candidate brief

The job is hybrid, with two days a week in the Manchester office. Fully remote working has not been agreed.

The hiring manager is still reviewing applications. This applicant has not been approved for an interview.

Applicants may contact the recruiting team to discuss assessment adjustments. Arrangements need to be agreed through the normal process.

No identity documents have been requested. Prepare a draft only. Do not send a message or change an application.

Now show the candidate this deliberately flawed draft:

Congratulations! You have been selected for interview. The role can be fully remote. Everyone must complete the same assessment without adjustments. Please reply with a copy of your passport so we can confirm your place.

Ask:

Check the draft against the brief. Prepare a corrected message for review, explain your important changes and identify anything you need to confirm before sending it.

What should a strong answer catch?

The draft contains four material errors:

1. The interview has not been approved.
2. Fully remote work has not been agreed.
3. Adjustments can be discussed.
4. No passport has been requested.

A corrected message could say:

Thank you for your interest in the role. The hiring manager is still reviewing applications, so we have not confirmed an interview yet.

The role includes two days a week in our Manchester office. Please contact our recruiting team if you would like to discuss an adjustment for a later assessment.

We will confirm the next step after the application review.

The wording does not need to match this sample. Check whether the candidate preserves the facts.

Then ask:

What would you confirm before sending this?

The brief gives no date for the next update. A good answer may suggest agreeing one with the team. The candidate should not invent a date and write it as though it were already promised.

Also ask:

Why did you remove the passport request?

This shows whether they understood the problem or simply shortened the email. The task gave no reason to request identity evidence.

Record the first answer before giving any prompt or hint.

Follow-up: change one fact and see what they update

Give the candidate this update:

The hiring manager has now approved an interview. No time has been agreed. The applicant has asked whether the role could be fully remote.

Only one decision has changed. The interview is approved. The working arrangement and appointment time remain unresolved.

The candidate might ask the manager:

Can you confirm whether fully remote working is possible for this role? The applicant has asked about it. The advertised arrangement is two days a week in Manchester.

A candidate-facing draft could say:

We'd like to invite you to interview. The role currently includes two days a week in Manchester. I'll ask the hiring manager whether fully remote working is possible.

Would you prefer to wait for that answer before we arrange the interview?

The exact wording can differ. Check whether the candidate keeps each decision in the correct state.

Exercise 2: use AI to write an update and handover

Give the candidate access to the approved AI tool and this brief:

It is Wednesday. A candidate's interview was interrupted. Their answers about previous experience were saved, but a question about handling a difficult customer was not completed.

The recruiter still needs to check whether the saved answers can be used. The candidate has been told not to restart while that review happens. Their application remains under consideration, and the relevant reminders have been paused.

Your team has promised an update by Thursday afternoon, even if the review is unfinished. No new interview time has been agreed.

You will be away on Thursday. A colleague may be able to cover, but has not yet agreed.

Use the assistant to prepare a candidate email and a handover request. Do not send anything. Explain what you checked before accepting the drafts.

The candidate has two jobs:

1. Write a useful update for the applicant.
2. Arrange ownership of the promised Thursday update.

What should the candidate email cover?

A sound draft could say:

I'm sorry your interview was interrupted. Your answers about previous experience were saved. Our recruiter still needs to check which answers we can use and what remains to be covered.

Please don't restart the interview while we check. Your application remains under consideration. We'll update you by Thursday afternoon, even if the review is still in progress.

The message should tell the applicant what happened, what to do now and when to expect an update.

It should not claim that the saved answers have already been accepted.

What should the handover cover?

A sound request could say:

Can you send this candidate's update by Thursday afternoon while I'm away? Please confirm today so I can arrange other cover if needed.

Their experience answers were saved, but the difficult-customer question was not completed. The recruiter still needs to check which answers can be used. We told the candidate not to restart; reminders are paused and no new interview is booked.

Please send the update even if the review is unfinished.

The colleague should be able to tell what remains unresolved, what they are being asked to do and that they need to confirm ownership.

Keep the internal handover out of the candidate email. The applicant should not have to solve your staffing problem.

Test whether they secure ownership

Tell the candidate:

You sent the handover request, but nobody replied. What would you do before leaving?

A strong answer recognises that the task is still unowned.

They might follow up, ask another colleague or escalate to the team lead. Sending a request does not complete a handover.

Then confirm that a colleague has agreed to cover. It is now Thursday and the review is still unfinished.

Ask what the colleague can tell the applicant.

A reasonable answer is:

The review of your saved answers is not finished yet. You don't need to restart the interview, and your application remains under consideration. I'm sorry to keep you waiting.

The team now needs to agree the next update date. The candidate can suggest one for confirmation. They should not invent one and present it as settled.

How should you score the exercises?

First ask the candidate to show which fact supports each important statement in their final drafts:

• application status
• saved answers
• restart instruction
• update deadline
• handover status

This quickly exposes unsupported claims, including turning “saved” answers into “accepted” answers or a requested handover into confirmed cover.

Use the same scorecard for every candidate:

Score the work you observed. Mark anything these exercises do not show as not assessed. Prompt length and AI vocabulary are not useful scores by themselves.

Keep the candidate's first answer separate from any correction made after a prompt. For example:

The applicant corrected the unapproved interview and remote-work claims independently. They left the passport request in the first draft. After being asked where the request came from, they removed it and explained why it was unnecessary.

This gives the hiring manager two useful pieces of evidence: what the candidate caught independently and what they could correct with help.

Before using the assessment with applicants, have two colleagues rate sample responses independently. Compare their reasoning and agree which mistakes matter, which deserve a follow-up and which skills can reasonably be taught.

Decide which gaps are teachable

Separate tool familiarity from judgement.

Needing help with an unfamiliar interface is different from accepting an invented interview approval. The first may need product training. The second is a judgement problem that matters directly to the work.

If training is realistic, name what the new hire would learn and who will check their early work. Singapore's AI Fluency Skills framework and the UK's AI Skills Boost are examples of broader training routes; choose support that matches the gap you actually found.

Finish with a short hiring-manager note that separates independent performance from help and training:

Observed independently: Produced an accurate update, checked the interview status and deadline, and did not claim the saved answers had been accepted.

Needed a prompt: [Record only if relevant.]

Teachable gap: [Name the specific tool or workflow skill the team could reasonably teach.]

This gives the manager evidence they can use in a decision, rather than a broad label such as “confident AI user”.

---

These are practice exercises to adapt with your hiring team. They are not validated measures of overall AI ability. Use the results alongside the other requirements of the role.